气候变化领域集成服务门户(CCPortal): Using deep learning to map retrogressive thaw slumps in the Beiluhe region (Tibetan Plateau) from CubeSat images

CCPortal

DOI	10.1016/j.rse.2019.111534
	Using deep learning to map retrogressive thaw slumps in the Beiluhe region (Tibetan Plateau) from CubeSat images
	Huang L.; Luo J.; Lin Z.; Niu F.; Liu L.
发表日期	2020
ISSN	00344257
卷号	237
英文摘要	Retrogressive thaw slumps (RTSs) are among the most dynamic landforms in permafrost areas, and their formation can be attributed to the thawing of ice-rich permafrost. The spatial distribution and impacts of RTSs on the Tibetan Plateau are poorly understood due to their remote location and the technical challenges of automatic mapping. In this study, we innovatively applied DeepLabv3+, a cutting-edge deep learning algorithm for semantic segmentation, to Planet CubeSat images, which are satellite images with high spatial and temporal resolution. Our method allows us to automatically delineate 220 RTSs within an area of 5200 km2 with an average precision of 0.541. The corresponding precision, recall, and F1 score are 0.863, 0.833, and 0.848 respectively, when the threshold of intersection over union is 0.5. Moreover, approximately 100 experiments on k-fold cross-validation (k = 3, 5, and 10) and data augmentation show that our method is robust. And a test in a different geographic area shows that the generalization of the trained model is very good. We find that (1) most of the RTSs are small (areas < eight ha and perimeters < 2000 m) and (2) RTSs preferentially develop at locations with gentle slopes (four to eight degrees), and in areas lower than the surroundings (the mean topographic position index is −0.17) and receiving less solar radiation (i.e., north-facing slopes). The results show that the method can map RTSs automatically from Planet CubeSat images and can potentially be applied to larger areas. © 2019 Elsevier Inc.
英文关键词	Convolutional neural network; Permafrost thawing; Planet CubeSat; Retrogressive thaw slumps; Tibetan plateau
语种	英语
scopus关键词	Edge detection; Image segmentation; Learning algorithms; Neural networks; Permafrost; Satellites; Semantics; Thawing; Convolutional neural network; K fold cross validations; Permafrost thawing; Semantic segmentation; Spatial and temporal resolutions; Thaw slump; Tibetan Plateau; Topographic positions; Deep learning; algorithm; artificial neural network; landform; permafrost; satellite imagery; solar radiation; spatial distribution; thawing; Qinghai-Xizang Plateau
来源期刊	Remote Sensing of Environment
来源机构	中国科学院西北生态环境资源研究院
文献类型	期刊论文
条目标识符	http://gcip.llas.ac.cn/handle/2XKMVOVA/179554
作者单位	Earth System Science Programme, Faculty of Science, The Chinese University of Hong Kong, Hong Kong SAR, Hong Kong; Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences, Lanzhou, China
推荐引用方式 GB/T 7714	Huang L.,Luo J.,Lin Z.,et al. Using deep learning to map retrogressive thaw slumps in the Beiluhe region (Tibetan Plateau) from CubeSat images[J]. 中国科学院西北生态环境资源研究院,2020,237.
APA	Huang L.,Luo J.,Lin Z.,Niu F.,&Liu L..(2020).Using deep learning to map retrogressive thaw slumps in the Beiluhe region (Tibetan Plateau) from CubeSat images.Remote Sensing of Environment,237.
MLA	Huang L.,et al."Using deep learning to map retrogressive thaw slumps in the Beiluhe region (Tibetan Plateau) from CubeSat images".Remote Sensing of Environment 237(2020).